Inconsistency of E-SNLI-VE dev\test between .csv and .json

[MyLittleChange]

Hi:
I've downloaded the e-SNLI-VE dataset, but i found that there are some label inconsistencies between the .json file and the .csv file. It seems like that .csv file uses the re-annotation label while .json file uses the former version. But the eUG code uses the .json file.
And it only occurs in dev anf test, dev has 1665 examples and est has 1631 examples that are inconsistent.

[maximek3]

Hey, thank you for raising this issue. It looks indeed like the e-SNLI-VE version we used (.json) mistakenly doesn't use the corrected neutral labels in the dev and test set. I uploaded the corrected version to the google drive link and will soon post updated results on e-SNLI-VE. Thanks for pointing this out and sorry for the inconvenience!